Electronic consumer products are pushing both the bounds of computation complexity and high clocking frequencies. The systems that are experiencing these problems include: VLSI (Very Large Scale Integration), microprocessors, ASIC's (Application Specific Integrated Circuit), SOC (System On a Chip) and FPGA's (Field Programmable Gate Arrays). All of these systems operate at clock frequencies that increase dramatically each year. This higher clock frequency coupled with the increased size of the die creates some fundamental problems in heat removal from the die and clock distribution over the surface of the die.
The power dissipation typically follows the rule:
                    P        =                              1            2                    ⁢                      CV            2                    ⁢          f                                    (        1        )            which for a 1 pF load being clocked at a frequency of 10 GHz is 7.2 mW. Adiabatic techniques can reduce these power dissipation levels. It would be worthwhile to investigate if these techniques can be used advantageously to help solve these problems.
Some of the basic circuit blocks to help achieve the ability for mobility, low power, and high computation require the necessity of a clock oscillator block. Tank circuits have been used to generate oscillatory clock signals. These circuits use LC (inductor-capacitor) elements to form the tank circuit and they have the adiabatic quality that may be used to reduce the power dissipation.
For example, U.S. Pat. No. 5,396,195 issued Mar. 7, 1995 to Gabara proposed a basic LC tank circuit in an MOS technology. The circuit consists of a tank circuit driven by a cross-coupled MOS circuit. The oscillations generated by the MOS LC tank circuit fabricated in a 0.9 um CMOS technology operated with a supply voltage of 3.3V. The power dissipation was reduced by a factor of a 10× when a capacitive load was driven using an LC tank circuit as compared to being driven using conventional digital techniques. This circuit has been used in a multitude of applications ranging from wireless to on-chip clock generation modules. Many of the inductors used in this type of tank circuit have the form of the horizontal planar inductor as illustrated in FIG. 1a and FIG. 1b. These type of inductors typically require a large amount of area to form the inductor.
The calculation of the values of planar inductors are provided in a published paper, “Simple Accurate Expressions for Planar Spiral Inductances”, IEEE J. Solid-State Circuits, Vol. 34, No. 10, October 1999, by Mohan et al., hereafter referred to as the “Mohan” reference.
In addition, the Q or quality factor of these inductors that are fabricated in CMOS are typically low. The quality factor or Q is a primary parameters in the evaluation of tank circuits.
                    Q        =                  2          ⁢          π          ⁢                                    Maximum              ⁢                                                          ⁢              energy              ⁢                                                          ⁢              stored              ⁢                                                          ⁢              in              ⁢                                                          ⁢              tank              ⁢                                                          ⁢              circuit                                      Energy              ⁢                                                          ⁢              dissipated              ⁢                                                          ⁢              per              ⁢                                                          ⁢              cycle                                                          (        2        )            
The Q indicates the amount of energy dissipated by the tank circuit to maintain oscillations. The tank circuit is more energy efficient as the value of the Q term increases which indicates that the energy dissipated in the tank circuit decreases. One way to decrease the dissipation is to reduce the parasitic resistance of the inductor.
An oscillator block provides the ability to regulate the flow of computation data within a VLSI (Very Large Scale Integration). For instance, the on chip clock frequency of a high-end microprocessor is expected to reach 10 GHz before the end of this decade. In addition, the power dissipation for the microprocessor is expected to be about 200 W, where the clock network will consume almost half of this power or 100 W. Thus, for this microprocessor, the higher frequencies and larger power dissipation values indicate a need to have clock circuits that can easily generate a 10 GHz signal and should be able to reduce the power dissipation of the clock network. The clock network of these VLSI chips typically contains large values of capacitance that need to be driven.
Currently, H-trees are used to distribute clocks over the surface of a die. Almost half of the power dissipated in chip designs occurs in the clock network of VLSI and microprocessor chips. This is largely due to the capacitive and resistive load of the clock network.
Several authors have addressed the clocking issue to determine achieve lower power, lower skew, and higher frequency of operation.
In O'Mahony et al., a U.S. PGPUB. 2003/0001652 A1 published Jan. 2, 2003, they use a hierarchical clock distribution. The clock is sent to a plurality of clock grids by way of transmissions lines, and then each grid distributes the clock to the load. They use salphasic clocking which takes advantage of standing waves along a transmission line. The position of the receiver points must conform to positions that are multiple of one-half wavelength from one another dependant on the clock frequency. This will lock the frequency of the die into a range dependant on the half-wavelength. The loads however do not have to obey this constraint.
In Galton et al., “Clock Distribution Using Coupled Oscillators”, Proceeding of the 1996 IEEE Inter. Symp. On Circuits and Systems, May 12-15, Vol. 3, pp. 217-220, they suggest using strongly coupled RC oscillators to distribute a clock signal over the die. Their technique uses transmission line that can be less than a quarter of a wavelength long. In addition, the transmission line can be lossy to couple the RC oscillators. Injection locking is used to lock all the oscillators in frequency.
In Hall et al., “Clock Distribution Using Cooperative Ring Oscillators”, Proc. IEEE 17th Conf. Advanced Research in VLSI, 1997, pp. 62-75, a cooperative ring oscillator is used to distribute a clock signal within the die. They also provide multiple clock phases in the distribution. Their circuit expands the ring oscillator from a simple ring to an N-dimensional mesh. This array does not have to be regular. One of the concerns is aggregation that is the non-ideal characteristic variations of the VLSI interconnect. This is a concern since interconnect that is used to connect the inverters of the ring oscillators. This factor dissipates power and introduces skew into the network.
In Wood et al., “Rotary Traveling-Wave Oscillator Arrays: A New Clock Technology”, IEEE J. Solid-State Circuits, Vol. 36, No. 11, November 2001, a rotary traveling wave oscillator array is presented. It consists of a balanced set of transmission lines with distributed CMOS latches to power the oscillation and ensure rotation lock. A waveform propagates along the balanced transmission line that is looped at its ends so that the wave continues to propagate. In their design it is important that careful attention is required to guard against magnetic field coupling between the clock conductors since it will affect the potential performance of their oscillators.
The technique presented here does not need to be constrained by half-wavelength or quarter-wavelength considerations as in O'Mahoney or Galton. In addition, Galton suffers from lossy transmission lines that are used to couple the RC oscillators together as well as the loss in the RC oscillator. As pointed out by Hall, the power dissipation in their technique is an issue for two factors; the ring oscillators dissipates power and the propagation of the signals in the interconnect dissipate power. The technique presented here uses adiabatic techniques to help overcome the particular losses of Hall and Galton. Finally, in Wood, magnetic field coupling is an undesirable condition; the technique presented here thrives on magnetic field coupling.
The need for frequency adjustment of an array of oscillators is an important factor to overcome the limitation of any clock distribution network that is based on a wavelength-based layout. The wavelength-based layout will have a limited range of tuning and it would be desirable to extend the tuning range of a distributed clock network. The technique presented here provides such an outlet.